News Exclusive: Unreleased Phison E28 SSD beats the fastest SSDs on the market in new benchmark

[Admin]

Tom's Hardware has acquired one of the first benchmarks for Phison's PS5028-E28 high-performance PCIe 5.0 SSD controller.

Exclusive: Unreleased Phison E28 SSD beats the fastest SSDs on the market in new benchmark : Read more

 

[Notton]

Fastest is all fine and dandy, but what about heat output and power consumption/efficiency?
We're finally seeing mobile chips with PCIe 5.0, so it's about time SSDs caught up with efficiency.

 

[S58_is_the_goat]

We need cheaper ssd's not faster ones.

 

[thestryker]

Fastest is all fine and dandy, but what about heat output and power consumption/efficiency?

This one will be on a more advanced node so compared to the rest of the high performance controllers it should be good. I imagine it will be closer to that of the high performance PCIe 4.0 controllers than current high performance PCIe 5.0.

I'd expect that the slower E31T, 4 channel, controller they released last year would be the one for laptop usage. That was their first PCIe 5.0 controller on a more advanced node (I believe one of the N7 ones) and is close to the lower power PCIe 4.0 controllers: https://www.tomshardware.com/pc-components/ssds/pny-cs2150-2tb-ssd-review/2

 

[emike09]

We need cheaper ssd's not faster ones.

And bigger SSDs, also for cheaper. PCIe 4.0 NVMe drives are plenty fast for what I do. I get that datacenter and some professional workloads could benefit, but the average consumer, even most professionals, don't need those speeds.

 

[Amdlova]

On my daily basis using a enterprise nvme pci-e 3.0 to an enterprise pci-e nvme 4.0 don't see any gains... Maybe it's windows thing. I will wait that new ones pci-e 5.0 drives comes down price.
I want to get an asus h770 board do finally do some optane raid 0 have two spare 280gb 900p here

 

[DougMcC]

That table .... Bytes per second? Is it maybe MB/s? That would still seem low but GB/s would seem too high.

 

[thestryker]

That table .... Bytes per second? Is it maybe MB/s? That would still seem low but GB/s would seem too high.

MB/s

 

[Mr Majestyk]

And bigger SSDs, also for cheaper. PCIe 4.0 NVMe drives are plenty fast for what I do. I get that datacenter and some professional workloads could benefit, but the average consumer, even most professionals, don't need those speeds.

Yes indeed, all I need is affordable 8TB drive for my photo library no need for PCI-E 5.0 at all. Unless we see a breakthrough in and that allows random performance approaching Optane, there is virtually no real world benefit to even going beyond PCI-E 3.0 unless you do sequential transfers all day for a living.

 

[DonQuixoteIII]

SSD's, like most things tech, need to be tested like for like to have any validity. This test is flawed by using a 4TB Samsung against 2TB for everyone else. Dr.Pabst must be laughing his ass off.

 

[VizzieTheViz]

We need cheaper ssd's not faster ones.

I’m all for faster ssd’s but I’d really really like for software to catch up and USE all that speed.

Apps and games with shorter startup/load times? Yes please!

As it is I notice no difference in my PCIe 5.0 T705 SSD and my NM970 PCIe 4.0 (and not really any difference with the PCIe 3.0 Samsung whatever its number was in my previous pc).

I want that experience I had with my first Intel postville sata ssd: really REALLY noticing my pc is a lot faster.

All the numbers keep going up on SSD’s but the experience using a pc doesn’t change a bit.

That said cheaper 8TB or bigger SSD’s for simple storage would be nice too.

 

[drinking12many]

I think we have gotten to the point most people can't tell the difference (short of enterprise etc). I noticed nothing noticeable between PCIe 3.0 and 4.0 in real terms. Sure benchmarks show a lot of difference, or internal copying etc, but I copy mostly between a NAS and my PC even at 10Gb there is no way to "notice" the speed. Game loads may be a few seconds faster but compared to the leap between even platters and SATA SSDs its just not as much of a WOW factor and honestly has me in no rush to go to PCIe 5.0 SSDs even though my current MB supports it, I just carried over my 4.0 drives.

 

[Penzi]

I’m all for faster ssd’s but I’d really really like for software to catch up and USE all that speed.

Apps and games with shorter startup/load times? Yes please!

As it is I notice no difference in my PCIe 5.0 T705 SSD and my NM970 PCIe 4.0 (and not really any difference with the PCIe 3.0 Samsung whatever its number was in my previous pc).

I want that experience I had with my first Intel postville sata ssd: really REALLY noticing my pc is a lot faster.

All the numbers keep going up on SSD’s but the experience using a pc doesn’t change a bit.

That said cheaper 8TB or bigger SSD’s for simple storage would be nice too.

I feel that way for most hardware. Why is so much software still single threaded? CPUs have been multi-core since 2006 (okay, that was the introduction, not when it became all multi-core).

Still, it shows that hardware is alive and well, offering regular performance enhancements and new features, so at least hardware isn’t stagnant!

And yes, I miss the old HDD days, at least for regular increases in capacity! Don’t really miss the speed, the bulk, and so on…

 

[DougMcC]

I feel that way for most hardware. Why is so much software still single threaded? CPUs have been multi-core since 2006 (okay, that was the introduction, not when it became all multi-core).

Still, it shows that hardware is alive and well, offering regular performance enhancements and new features, so at least hardware isn’t stagnant!

And yes, I miss the old HDD days, at least for regular increases in capacity! Don’t really miss the speed, the bulk, and so on…

Multithreading is surprisingly hard. The percentage of software devs who can really deliver big performance gains with it is about 1%.

 

[BFG-9000]

CPUs have been multi-core since 2006 (okay, that was the introduction, not when it became all multi-core).

Both Smithfield Pentium D and Manchester A64 X2 were released in early 2005, with the regular Pentium D first by 5 days (+ Extreme Edition a month earlier), whereupon AMD claimed to have the first "true" dual-core because theirs was monolithic die while Intel just slapped two Prescott Pentium 4 dies onto the same package, nevermind that neither chip shared L2 cache between cores. Intel had used two virtual cores thanks to Hyperthreading since 2002, but only the Extreme Edition Pentium Ds had Hyperthreading for 4 virtual cores. Core 2 came out in 2006 and dispensed with Hyperthreading as it wasn't that useful without the 31-flavor long pipeline of Prescott, and aside from the rare Core 2 Solo was almost entirely multicore.

I should point out that the main memory speeds of these two platforms was only 10.67GB/s and 6.4GB/s respectively, on DDR2-667 or DDR400 when you look at the speeds of these SSDs. Yes I am old.

 

[emike09]

Both Smithfield Pentium D and Manchester A64 X2 were released in early 2005, with the regular Pentium D first by 5 days (+ Extreme Edition a month earlier), whereupon AMD claimed to have the first "true" dual-core because theirs was monolithic die while Intel just slapped two Prescott Pentium 4 dies onto the same package, nevermind that neither chip shared L2 cache between cores. Intel had used two virtual cores thanks to Hyperthreading since 2002, but only the Extreme Edition Pentium Ds had Hyperthreading for 4 virtual cores. Core 2 came out in 2006 and dispensed with Hyperthreading as it wasn't that useful without the 31-flavor long pipeline of Prescott, and aside from the rare Core 2 Solo was almost entirely multicore.

I should point out that the main memory speeds of these two platforms was only 10.67GB/s and 6.4GB/s respectively, on DDR2-667 or DDR400 when you look at the speeds of these SSDs. Yes I am old.

Dang. I wasn't benchmarking back then, but 10.6GB/s doesn't sound bad. I'm currently on a G.Skill DDR5 kit running 6000MT/s, maxing at CL26-34-32-72 and I get 87GB/s read, 92GB/s write @ 65ns with manually tuned sub-timings and raised voltage. ~8x performance in 20 years doesn't sound impressive.

My super stable CL28 specs:
https://preview.redd.it/timings-for...bp&s=203da1c54818d11ec1d3a93f27c92a23d76cd029

 

[BFG-9000]

~8x performance in 20 years doesn't sound impressive.

If you think that's bad, DDR5-6000 at CL36 has a latency of 12ns, which is exactly the same as Jedec standard CL4 DDR2-667 while CL2.5 DDR-400 is 12.5ns. Your DDR5-6000 at CL26 is 9ns while DDR2-1066 at CL4 was 7.5ns or standard CL5 9.375ns, while standard DDR2-800 CL4 and DDR-400 at CL2 were both 10ns.

So while bandwidth has improved by 8x, latency has pretty much not improved at all. While the clock has increased, so has the CAS which is the number of memory clocks to wait before access, in your case 26. That's why CPUs have so much cache nowadays.

 

[Penzi]

Multithreading is surprisingly hard. The percentage of software devs who can really deliver big performance gains with it is about 1%.

That’s relevant but I don’t think that’s it. I assume it’s a business decision with low hanging fruit. Business as usual is cheaper even if your employees are capable of changing underlying code to take advantage of parallelization. Software tends to have lock-in and thus there is no necessary competitive advantage to superior code. To be clear this is only a generalization with all attendant caveats!

 

[CRamseyer]

You will see a much larger performance increase with E28 than with our previous flagship Gen5 SSD. We haven't optimized the low-queue-depth random read performance yet, at least on the drives displayed publicly or in any of the screenshots that leaked from a tester.

Our competitors build their next-gen products to compete with E26, and the reviews show those products only reach parity with E26. Our goal is to elevate E28 to the next level and deliver a performance boost where it matters. At Computex, we showed 32 drives in a be Quiet! case without a heatsink on a single E28 SSD. The system also featured an NVIDIA Ada-generation workstation GPU and an AMD Threadripper processor, accompanied by AMD EXPO memory with eight sticks running at 6400 MHz. Even with all of that processing power, an enormous number of drives, and components known for pushing thermal boundaries, we managed to reach the Windows 11 kernel limit of 114GB/s read and 104GB/s write speeds to disk.

E28 is the Gen5 SSD we've all dreamed of having.

 

[williamjeremiah]

We need cheaper ssd's not faster ones.

As usual, less costly hardware will eventually be the result, as faster hardware becomes available.